Software MAGiC para RM.

Prévia do material em texto

ORIGINAL RESEARCH
ADULT BRAIN
Synthetic MRI for Clinical Neuroimaging: Results of the
Magnetic Resonance Image Compilation (MAGiC) Prospective,
Multicenter, Multireader Trial
X L.N. Tanenbaum, X A.J. Tsiouris, X A.N. Johnson, X T.P. Naidich, X M.C. DeLano, X E.R. Melhem, X P. Quarterman,
X S.X. Parameswaran, X A. Shankaranarayanan, X M. Goyen, and X A.S. Field
ABSTRACT
BACKGROUNDANDPURPOSE: SyntheticMR imaging enables reconstruction of various image contrasts from 1 scan, reducing scan times
and potentially providing novel information. This study is the first large, prospective comparison of synthetic-versus-conventional MR
imaging for routine neuroimaging.
MATERIALS ANDMETHODS: A prospective multireader, multicase noninferiority trial of 1526 images read by 7 blinded neuroradiologists
was performed with prospectively acquired synthetic and conventional brain MR imaging case-control pairs from 109 subjects (mean,
53.0� 18.5 years of age; range, 19–89 years of age) with neuroimaging indications. Each case included conventional T1- and T2-weighted,
T1 and T2 FLAIR, and STIR and/or proton density and synthetic reconstructions frommultiple-dynamicmultiple-echo imaging. Imageswere
randomized and independently assessed for diagnostic quality, morphologic legibility, radiologic findings indicative of diagnosis, and
artifacts.
RESULTS: Clinical MR imaging studies revealed 46 healthy and 63 pathologic cases. Overall diagnostic quality of synthetic MR images was
noninferior to conventional imaging on a 5-level Likert scale (P� .001; mean synthetic-conventional,�0.335� 0.352; �� 0.5; lower limit
of the 95% CI,�0.402). Legibility of synthetic and conventional morphology agreed in�95%, except in the posterior limb of the internal
capsule for T1, T1 FLAIR, and proton-density views (all,�80%). Synthetic T2 FLAIR hadmore pronounced artifacts, including�24.1% of cases
with flow artifacts and�17.6% cases with white noise artifacts.
CONCLUSIONS: Overall synthetic MR imaging quality was similar to that of conventional proton-density, STIR, and T1- and T2-weighted
contrast views across neurologic conditions.While artifactsweremore common in synthetic T2 FLAIR, thesewere readily recognizable and
did not mimic pathology but could necessitate additional conventional T2 FLAIR to confirm the diagnosis.
ABBREVIATIONS: �� Kappa statistic; MAGiC�MAGnetic resonance image Compilation; MDME� multiple-dynamic multiple-echo; PD� proton density
Synthetic MR imaging uses quantitative probing of multiplephysical properties to reconstruct multiple contrasts from 1
scan. Parameters like TR, TE, and TI can be modified with math-
ematic inferences rather than being predetermined.1-3 The speed
of diagnostic brain studies can thus be reduced to only about 5
minuteswith syntheticMR imaging.4 This advancementmayhelp
improve throughput and reduce rescanning, while also providing
quantitative information of research interest.4-6
Clinical studies of synthetic MR imaging are highly hetero-
geneous in that they examine a variety of conditions with
widely varying scan parameters, with a paucity of large, ran-
domized trials to inform clinical usage.3,6 Blystad et al5 (2012)
reported that synthetic images had diagnostic utility similar to
that of conventional imaging series, though with some quality
issues like granulation and contrast particularly apparent in
FLAIR views. Other studies reported good quality and contrast
Received December 2, 2016; accepted after revision February 2, 2017.
From Lenox Hill Radiology (L.N.T.), RadNet Inc, New York, New York; Department
of Radiology (A.J.T.), Weill Cornell Medical Center, New York, New York; Depart-
ment of Technical Communication (A.N.J.), Science and Healthcare, Texas Tech
University, Lubbock, Texas; Technology and Medical Innovation Organization
(A.N.J., S.X.P.), Healthcare Imaging-MRI (P.Q.), Healthcare Imaging-MRI Neuro Appli-
cations (A.S.), and Medical Affairs (M.G.), GE Healthcare, Milwaukee, Wisconsin;
Department of Neuroradiology (T.P.N.), The Mount Sinai Hospital, New York, New
York; Division of Radiology and Biomedical Imaging (M.C.D.), Michigan State Uni-
versity, Advanced Radiology Services, PC, and Spectrum Health, Grand Rapids,
Michigan; Department of Diagnostic Radiology and Nuclear Medicine (E.R.M.), Uni-
versity of Maryland School of Medicine, Baltimore, Maryland; and Department of
Radiology (A.S.F.), University of Wisconsin School of Medicine and Public Health,
Madison, Wisconsin.
This work was supported by GE Healthcare to participating institutions.
Please address correspondence to Lawrence Tanenbaum, MD, 400 E 66th St, NY,
NY 10065; e-mail: Nuromri@radnet.com; @nuromri
Indicates open access to non-subscribers at www.ajnr.org
Indicates article with supplemental on-line tables.
Indicates article with supplemental on-line photos.
http://dx.doi.org/10.3174/ajnr.A5227
AJNR Am J Neuroradiol ●:● ● 2017 www.ajnr.org 1
 Published April 27, 2017 as 10.3174/ajnr.A5227
 Copyright 2017 by American Society of Neuroradiology.
for synthetic images among certain indications, such as multi-
ple sclerosis, brain metastasis,6,7 and myelination patterns.8
Because image quality endpoints are reliant on reader judg-
ment (reported to have up to 41% variability and only fair-to-
moderate interrater agreement9,10) and scanning conditions,
drawing clinically relevant inferences from diverse small trials
is challenging. Furthermore, the broad diversity of both
healthy and pathologic morphologic variants encountered in
routine neuroimaging necessitates more robust clinical studies
of synthetic MR imaging for clinical neuroimaging.
This study was designed to compare the overall image qual-
ity of synthetic MR imaging with conventional MR imaging in
a general neuroimaging population. Secondary aims included
legibility of anatomic and morphologic features, artifact prev-
alence, and diagnostic performance across a range of cases
helpful in informing clinical usage and adoption of synthetic
MR imaging.
MATERIALS AND METHODS
Participants and Clinical Assessments
Subjects (n � 117) were enrolled prospectively into a multi-
reader multicenter case-control study across 6 hospitals from
November 2015 to January 2016 (ClinicalTrials.gov Identifier
NCT02596854). Of these, all complete cases (n� 109; 45 men,
64 women; mean, 53.0� 18.5 years of age; range, 19–89 years)
with synthetic and conventional (control) acquisitions were
read. Subjects were 18 years of age or older with clinical indi-
cations for neuroimaging and without contraindications to
MR imaging or previously diagnosed congenital conditions or
extensive trauma prohibiting scanning. Governing ethics com-
mittees at each site approved this study, and subjects provided
written informed consent.
Image Acquisition
Images were prospectively acquired by using a fixed set of scan-
ning parameters closely approximating current standard of care
brain MR imaging (as detailed for 1.5T and 3T scanners in On-
line Table 1). First, conventional images were acquired by using
conventional 2D axial plane T1- and T2-weighted, T1 and T2
fluid-attenuated inversion recovery, short tau inversion recovery,
and proton density (PD) sequences. Then, a multiple-dynamic
multiple-echo (MDME) sequence was performed for synthetic
reconstruction, for a complete conventional and synthetic case-
control series. MDME uses a repeat version of the same gradient-
reversal process used to create a single gradient-echo to produce
additional gradient-echoes after a single radiofrequency pulse.
This is known as multiple (or dual) echo gradient-echo, which is
possible when complete loss of the transverse magnetization by
T2* relaxation has not yet occurred. Because MDME is a quanti-
tative sequence, it enablesabsolute quantification of tissue phys-
ical properties, like longitudinal R1 relaxation rate, transverse R2
relaxation rate, and PD independent of the scanner settings.
MDME parameters acquired in 1 scan are used in synthetic imag-
ing to calculate pixel intensity, producing an appearance similar
to that of conventional MR images with modifiable TE, TR, and
TI.5,11 Thus, synthetic (based on MDME) and conventional T1,
T2, T1 FLAIR, T2 FLAIR, PD, and STIR contrast views were col-
lected. MDME data were reconstructed outside the clinical care
environment by using MAGnetic resonance image Compilation
(MAGiC) software on a 64-bit Advantage Workstation (GE
Healthcare, Milwaukee, Wisconsin). No errors were logged dur-
ing processing, and the average processing time was approxi-
mately 2minutes per case. Scan duration, subject disposition, and
imaging results were recorded for each case.
The site-determined diagnosis was recorded on the basis of
the results of MR imaging studies and work-up performed ac-
cording to the standard of care by clinical neuroradiologists.
The sites reported the reference (site-determined diagnosis) by
using the same scale as the study readers, which reports normal
or�1 pathologic subtype adapted from Osborn et al (2010)12:
1) traumatic, complex, indeterminate, or other condition or
injury; 2) congenital malformation; 3) ischemic or hemor-
rhagic stroke; subarachnoid hemorrhage/aneurysm; 4) vascu-
lar malformation; 5) neoplasm/primary neoplastic cysts; 6) in-
fectious/demyelinating disease; or 7) metabolic/degenerative
disorders.
Radiologic Assessments
Synthetic and conventional images sets were randomized and
assessed by 7 blinded independent neuroradiologists (�10
years’ experience) on standard imaging workstations. Case-
control pairs from the same subject were separated and read
across 2 sessions, separated by a 4-week memory-washout pe-
riod. Each read included either all synthetic or all conventional
contrast views from a case. Overall diagnostic image quality
was rated (considering all available contrast views) on a
5-point Likert-type scale: 5 � excellent (acceptable for diag-
nostic use), 4 � good (acceptable for diagnostic use), 3 �
acceptable (acceptable for diagnostic use but with minor is-
sues), 2 � poor (not acceptable for diagnostic use), or 1 �
unacceptable (not acceptable for diagnostic use). Ratings of
�3 were considered acceptable overall. For image sets rated as
unacceptable (1 or 2), the rationale was recorded as “open
text.” Readers also recorded radiologic findings indicative of a
diagnosis with corresponding Osborn classifications.
For each contrast view, readers rated the legibility (or visibility
of margins and structures associated with key anatomic/morpho-
logic features) of anatomies defined a priori. Legibility ratings
supplemented overall image-quality data, which consider all re-
gions of the brain, as a means of providing specific information
about anatomic regions in brain imaging. Each anatomywas rated
on a binary scale (legible/illegible), including the following: cen-
tral sulcus, head of the caudate nucleus, posterior limb of the
internal capsule, cerebral peduncle, middle cerebellar peduncle,
and cervicomedullary junction. Readers recorded whether any of
the following artifacts were present:13 low signal-to-noise,motion
and section issues, infolding or wrap-around, white pixel or spike
noise, phase encoding, flow, contrast-to-noise, low image resolu-
tion, or blurring. Readers could provide free text comments on
any other observations.
Statistical Analysis
Statistical analysis was performed in SAS 9.2 (SAS Institute, Cary,
North Carolina), and sample size was calculated in PASS12
2 Tanenbaum ● 2017 www.ajnr.org
(NCSS Statistical Software, Kaysville, Utah). Per the prospective
statistical plan to determine noninferiority, a Wilcoxon signed
rank test was used to determine noninferiority of synthetic-to-
conventionalMR imaging in terms of the overall diagnostic image
quality score, by using a 1-sided �� .025 test with a noninferior-
ity margin of � � .5 with a 5-level Likert scale. The primary
hypothesis is 1-sided and can be stated asH0:S��� andHA:S�
��, where the S is the median difference of overall diagnostic
image quality across readers for synthetic-versus-conventional
MR imaging, in which noninferiority is established by rejecting
the null hypothesis. The margin (�) of .5 was determined sta-
tistically on the basis of the population and was confirmed by
clinical estimates from prior research5 and institutional pilot
data, in accordance with recommendations for determination
of noninferiority margins described in the US Food and Drug
Administration Guidance for Industry: Non-Inferiority Clinical
Trials to Establish Effectiveness (2016) (https://www.fda.gov/
downloads/Drugs/GuidanceComplianceRegulatoryInformation/
Guidances/UCM202140.pdf) and trial designs for noninferiority
testing in radiology reviewed by Ahn et al (2012).14 Descriptive sta-
tistics were used to summarize secondary endpoints of anatomic/
morphology legibility by anatomic region, artifact prevalence, and
diagnostic performance (sensitivity/specificity) by theOsborn classi-
fication. Interrater reliability between readers was assessed by kappa
(�) statistic.
RESULTS
Overall Diagnostic Quality
Each of 7 blinded neuroradiologists read all 109 clinically ac-
quired case-control image sets (109 synthetic and 109 conven-
tional) for a total of 1526 reads (763 synthetic and 763 conven-
tional reads). Of these, 56/109 were acquired on 1.5T static field
strength scanners and 53/109 were acquired on 3T scanners. Be-
cause no significant differences for 5-level image quality scores
(acceptable � 3, 4, or 5 versus unacceptable � 1 or 2) were ob-
served on the basis of scanner static field strength (1.5T or T) or
acquisition site (P� .05 with a 2-tailed t test), results were pooled
for analysis. The duration of scanningwas recorded, with a single-
acquisition sequence for synthetic reconstruction requiring 5
minutes 36 seconds on 1.5T scanners and 5 minutes 4 seconds on
3T scanners (On-line Table 1).
Considering all contrast views, 734 (96%) synthetic cases and
745 (98%) conventional cases were rated as acceptable (�3 on a
5-point scale) (Table). Figure 1 shows comparable synthetic and
conventional case-control images from a normal (no pathology
present) brain by contrast view. Figures 2–5 show case-control
Diagnostic image-quality ratings by static field strength of scanner and overalla
Diagnostic Qualityb
1.5T Scanner 3T Scanner Overall (1.5T + 3T)
Syn (N = 392) Con (N = 392) Syn (N = 371) Con (N = 371) Syn (N = 763) Con (N = 763)
Acceptable for diagnostic use (3,4,5) 380 (97%) 387 (99%) 354 (95%) 358 (96%) 734 (96%) 745 (98%)
Excellent (rated 5) 64 (16%) 183 (47%) 37 (10%) 103 (28%) 101 (13%) 286 (37%)
Good (rated 4) 266 (68%) 170 (43%) 230 (62%) 199 (54%) 496 (65%) 369 (48%)
Acceptable (rated 3) 50 (13%) 34 (9%) 87 (23%) 56 (15%) 137 (18%) 90 (12%)
Unacceptable for diagnostic use (1,2) 12 (3%) 5 (1%) 17 (5%) 13 (4%) 29 (4%) 18 (2%)
Poor (rated 2) 10 (3%) 5 (1%) 16 (4%) 12 (3%) 26 (3%) 17 (2%)
Unacceptable (rated 1) 2 (1%) 0 1 (0.3%) 1 (0.3%) 3 (0.4%) 1 (0.1%)
Note:—Syn indicates synthetic MR imaging; Con� conventional MR imaging (control).
a All data are shown as n (n/N%), where n is the count and N is the total reads per category (defined in the upper row of this table).
b Five-point scale.
FIG 1. Axial synthetic and conventional 3TMR imaging of a normal brain. Conventional (upper row) and synthetic (lower row) image sets exhibit
similar legibility and quality. Slight differences in contrast levels are apparent, which do not adversely impactthe diagnostic utility of images,
particularly between T1 FLAIR and T2 FLAIR views.
AJNR Am J Neuroradiol ●:● ● 2017 www.ajnr.org 3
images across a range of brain pathologies (continued in On-line
Figs 1–3). Overall diagnostic image quality of synthetic images
was statistically noninferior to conventional images, with a mean
difference (synthetic-conventional) across readers of �0.335 �
0.352 with a lower limit of the (1-sided) 95% CI of�0.402 (me-
dian, �0.428; minimum, �1.286; and maximum, 0.714; P �
.001). Among synthetic images rated as poor or unacceptable (1
or 2 on a 5-point scale), the most common quality issue was pa-
tient motion in synthetic image sets owing to generating from a
single acquisition (where a single motion event propagates across
all reconstructed contrast views).
Legibility of Anatomic/Morphologic Features
Anatomic/morphologic features were visualized and rated as leg-
ible in synthetic and conventional imaging for �98% of regions
across contrast views, except in the cervicomedullary junction
rated at 96% on both synthetic and conventional imaging (On-
line Table 2). For synthetic and conventional pairs from the same
subject, readers agreed for �95% of anatomic/morphologic re-
gions across contrast views, except in the posterior limb of the
internal capsule for T1, T1 FLAIR, and PD views (�80% agree-
ment). Notably, 6 of 7 readers had agreement of 99%–100% for
T1 FLAIR, with 1 reader as an outlier at 89%, possibly related to
experience. Further study will be needed to investigate the in-
fluence of experience on reading synthetic images and possible
training solutions.
Artifacts Occurrence and Characterization
Fewer artifacts (all characterizations) were identified in synthetic
than in conventional imaging for T1-weighted (9.2%), STIR
(24.8%), and PD (1.1%) contrast views (On-line Table 3). Syn-
thetic images had more artifacts overall on the T2-weighted
(5.0%), T1 FLAIR (17.9%), and T2 FLAIR (49.3%) contrast views
(On-line Table 3). Phase-encoding artifacts were less frequent in
synthetic STIR images (27.2%) and synthetic T1 contrast views
(13.0%). Synthetic contrast views were more likely to contain
FIG 2. Left frontal lobe cystic tumor on axial synthetic and conventional 3T MR imaging in a 31-year-old woman. Conventional (upper row) and
synthetic (lower row) image sets exhibit similar legibility and quality.
FIG 3. Chronic right cerebellar infarction in a 37-year-old woman on axial synthetic and conventional 3T MR imaging. Conventional (upper row)
and synthetic (lower row) image sets exhibit similar legibility and quality.
4 Tanenbaum ● 2017 www.ajnr.org
white pixels/spike noise artifacts across contrast views (except
PD), and flow artifacts were more common in synthetic views,
most notably in the synthetic T1-weighted (13.0%), T1 FLAIR
(22.1%), and T2 FLAIR (24.1%) contrast views.
Readers identified relatively more artifacts among synthetic
T2 FLAIR contrast views compared with other synthetic and con-
ventional contrast views. Synthetic T2 FLAIR showed 24.1%more
flow artifacts, 17.6% more white noise artifacts, and 59.2% more
artifacts marked as “other” compared with conventional views.
Examination of reader free text comments revealed that artifacts
marked as “other” primarily described localized, granulated hy-
perintensities apparent in themargins only in synthetic T2 FLAIR
contrast views (Fig 6). These artifacts were recognizable by a dis-
tinct pixelated appearance and a tendency to occur along tissue-
CSF boundaries only in T2 FLAIR views in otherwise unremark-
able image sets. Readers reported that synthetic T2 FLAIR may
have some diagnostic limitations in practice, which could neces-
sitate a conventional T2 FLAIR scan. However, owing to the na-
ture of synthetic imaging (which results in a full range of possible
contrast views for cross-comparison), neuroradiologists were
readily able to distinguish T2 FLAIR artifacts from pathology,
without impacting diagnostic utility.
Diagnostic Performance
Overall interrater agreement (� correlation coefficient) for pa-
thology detection was 0.502 for synthetic images and slightly
higher at 0.605 for conventional images. Across the 7 readers
(1526 total reads, including 763 synthetic and 763 conventional
pairs), overall sensitivity for correct identification of pathology
ranged from 60.32 (95% CI, 47.20–72.43) to 95.24 (95% CI,
86.71–99.01) among readers for conventional versus 55.56 (95%
CI, 42.49–68.08) to 96.83 (95% CI, 89.00–99.61) among readers
for synthetic imaging, with the site-determined diagnosis as the
criterion standard. On the basis of clinically confirmed diagnoses
reported by the site (based on clinical MR imaging studies and,
when necessary, additional follow-up or laboratory testing), the
study included 46 healthy and 63 pathologic cases (of which 2
cases contained 2 pathology types and 1 case contained 3 pathol-
ogy types), including 7 traumatic or complex injuries, 2 congen-
ital malformations, 12 strokes/hemorrhages, 2 vascular malfor-
mations, 32 neoplasms/primary neoplastic cysts, 10 infectious/
demyelinating conditions, and 2 metabolic/degenerative disorders.
Readers of synthetic MR imaging showed equal or higher ability to
diagnose all pathologies, except for neoplasms/primary neoplastic
cysts (n� 2, difference in detection of�6.3% sensitivity and�1.3%
specificity among readers) subgroup and infectious diseases (n� 10,
difference indetectionof�10.0%sensitivity and�3.0%specificity).
DISCUSSION
To our knowledge, this is the first large, prospective, randomized
study of synthetic MR imaging technology to enroll a cross-sec-
tion of the neuroimaging population, including a variety of brain
pathologies encountered in clinical practice. On the basis of
blinded assessments from 7 neuroradiologists, the overall diag-
nostic quality of synthetic MR imaging was statistically noninfe-
rior to conventional MR imaging series for T1- and T2-weighted,
T1 and T2 FLAIR, STIR, and PD contrast views. Furthermore,
neuroradiologists reported similar anatomic/morphologic fea-
ture legibility in both synthetic and conventional images. Both
synthetic and conventional sequences exhibited similar quality
issues and artifact trends for T1- and T2-weighted, STIR, and PD
contrast views, while synthetic imaging hadmore FLAIR artifacts.
Synthetic FLAIR artifacts were readily recognizable by cross-com-
parison within contrast views and thus did not significantly im-
pact the diagnostic use of synthetic MR imaging. Overall, study
results demonstrated that both synthetic and conventional imag-
ing have similar diagnostic utility.
Anatomic and morphologic characteristics were visible in
FIG 4. Multiple sclerosis in axial synthetic and conventional 3T MR imaging in a 58-year-old woman. Multifocal demyelinating lesions are
apparent within the cerebral white matter; these lesions appear similar on conventional (upper row) and synthetic (lower row) image sets. A
slight misregistration is apparent due to patient motion between the MDME scan (used for synthetic reconstruction, lower row) and the
comparable conventional scan acquired in the study (upper row).Whilemisregistration due tomotion can pose challenges in conventional serial
acquisitions due to partial section differences in images across contrast views, synthetic reconstruction inherently prevents misregistration
across synthetic contrast views.
AJNR Am J Neuroradiol ●:● ● 2017 www.ajnr.org 5
both synthetic and conventional views, though both exhibited
issues pertaining to visualizing the craniocervical junction, CSF
suppression, and pulsation artifacts that are well-documented in
MR imaging.5,15-18 Synthetic imaging exhibited characteristichy-
perintense artifacts in FLAIR views, corroborating previous re-
ports that further work will be necessary before synthetically gen-
erated FLAIR views can fully replace conventional FLAIR in
practice.5,6 While FLAIR artifacts contributed to lower overall
image quality scores (because all views were considered in this
composite primary end point), the overall impact of FLAIR arti-
facts on diagnosis was inherently limited by the nature of the
synthetic views, inwhich immediate cross-comparisonwith other
contrast views is possible. On rare occasions, encoding artifacts in
FLAIR views could necessitate clinical workflow changes such as
the addition of a single conventional scan; however, the impact
on the patient’s overall scan experience is offset by the time sav-
ings of the synthetic acquisition. Furthermore,motion and signal-
encoding artifacts were observed to affect all reconstructed syn-
thetic views if present in the original acquisition. As few as 7.5%of
singleMR images exhibitedmotion artifacts, while up to 19.8%of
long scans of multiple contrasts may be
affected.19 Because synthetic imaging re-
duces the overall scan time, the impact
of acquisition issues is expected to be
limited in practice.
Diagnostic performance of synthetic
imaging was similar to that of conven-
tional MR imaging, as indicated by sta-
tistical noninferiority of synthetic im-
ages. While the noninferiority model is
decisive for effectiveness in therapeutic
studies, which directly assess ultimate
patient outcomes, elucidating the clini-
cal implications of noninferiority find-
ings in radiology is less straightforward
because the negative effects of image
quality may have variable effects on ulti-
mate patient outcomes.14 Thus, from a
clinical perspective, we observed that in
both synthetic and conventional MR
imaging, some neoplasms/primary neo-
plastic cysts and infectious or demyeli-
nating conditions were challenging for
readers to identify without additional
clinical or laboratory work-up, possibly
due to overlapping appearances of neo-
plastic and inflammatory conditions on
MR images.20,21
The sensitivity and specificity of MR
imaging in neuroradiology have been
reported to range from 39% to 98% and
33% to 100%, respectively, with wide
variations based on reader experience
and the pathologic condition stud-
ied.22-25 Across study readers, synthetic
MR imaging sensitivity and specificity
had values within typical clinically ob-
served ranges for blinded MR imaging
reads (without clinical context).22-25 Statistical variations in diag-
nostic classifications may be centrally attributable to small sam-
ples of certain pathologies in the present study, meriting further
study of these pathologic subgroups. Synthetic scanning is per-
formed in the axial view only, and some clinical cases may be
limited by spatial resolution in this section direction. Owing to
the relatively shorter synthetic acquisition time, however, ad-
ditional sequences can also be combined with the synthetic
acquisition in a single examination session with minimal bur-
den on the patient.
The strengths of this study include the use of a prospective
acquisition protocol withmatched scanning parameters (On-line
Table 1). Because scans were acquired in a fixed order with
MDME (synthetic reconstruction) acquired last, a relative pro-
pensity toward motion artifacts in synthetic images may not be
representative of actual occurrence. Reports have, however,
shown that single scans of short duration have lower incidences of
motion than longer scans.13,26 The trial results support the use of
synthetic MR imaging in brain imaging to reduce scan time and
FIG 5. Chronic infarction in synthetic and conventional 3T MR imaging shown alongside color
functional perfusion maps in a 62-year-old man. Conventional (rows 1 and 3) and synthetic (rows
2 and 4) views show similar legibility and quality. T1 FLAIR and T2 FLAIR views have some granu-
lated white noise in the margins. Color quantitative perfusion maps (lower right) demonstrate
decreased flow and prolonged transit time in this region of chronic infarction.
6 Tanenbaum ● 2017 www.ajnr.org
the associated discomfort for patients undergoing brain MR im-
aging, with diagnostic performance similar to that of conven-
tional imaging.
CONCLUSIONS
The current study demonstrated that synthetic images were sta-
tistically noninferior in terms of overall diagnostic image quality
compared with conventional MR images, with similar diagnostic
utility for detecting a range of brain pathologies. Both synthetic
and conventionalMR imaging could visualize anatomic andmor-
phologic features of the brain, with similar trends in artifacts and
diagnostic utility. Because synthetic reconstructions rely on the
quality of a single scan, care should be taken to minimize motion
and acquisition artifacts. While more artifacts were observed in
synthetic T2 FLAIR reconstructions, cross-comparison with
other contrast views enabled neuroradiologists to readily detect
these artifacts without interfering with the diagnostic ability of
synthetic images. The trial results support the use of syntheticMR
imaging in brain imaging to reduce scan time and discomfort for
patients undergoing brain MR imaging, while acquiring high-
quality diagnostic MR images. We expect that further research
may reveal additional applications for synthetic MR imaging.
ACKNOWLEDGMENTS
We thank the participants of the study, the study coordinators
across multiple sites, and the study readers. We also acknowledge
the contributions of DrMichael Starc (Department of Radiology,
Weill Cornell Medical College, New York, NY) in study conduct
and of Dr Keyi Wang (GE Healthcare, Technology and Medical
Innovation) for assistance with the statistical design of the trial.
Disclosures: Lawrence N. Tanenbaum—RELATED:Grant: GE Healthcare, Comments:
sponsored trial payments for patient scanning*;UNRELATED:Consulting Fee orHon-
orarium: GE Healthcare;Consultancy:GEHealthcare. Angela N. Johnson—RELATED:
Support for Travel to Meetings for the Study or Other Purposes: GE Healthcare,
Comments: Funding for travel was provided by GE Healthcare for US FDA agency
and investigator meetings associated with study conduct. No additional funding was
provided for the publication activities; UNRELATED: Employment: GE Healthcare,
Comments: Angela N. Johnson is an employee of the Technology and Medical Inno-
vation Organization of GE Healthcare. Providing publication assistance is not a re-
quirement of employment or incentivized by GE Healthcare; Payment for Lectures
Including Service on Speakers Bureaus: Regulatory Affairs Professional Society,
Comments: conference faculty for the annual conference, receiving honorarium;
Stock/Stock Options: GE Healthcare, Comments: I hold employee stock options in
GE Healthcare; OTHER RELATIONSHIPS: I am a doctoral student at Texas Tech Uni-
versity receiving employee-sponsored tuition benefits for doctoral research. Ajit
Shankaranarayanan—UNRELATED: Employment: GE Healthcare. Mathias Goyen—
UNRELATED: Employment: GE Healthcare, Comments: full-time employment, Chief
Medical Officer Oncology; Stock/Stock Options: GE Healthcare. Aaron S. Field—
RELATED: Grant: GE Healthcare, Comments: grant from study sponsor to cover
institutional costs associated with participation in the trial. This did not include any
salary or other support paid to me*. *Money paid to the institution.
REFERENCES
1. Riederer S, Suddarth S, Bobman S. Automated MR image synthesis:
feasibility studies. Radiology 1984;153:203–06 CrossRef Medline
2. Warntjes J,DahlqvistO, LundbergP.Novelmethod for rapid, simul-
taneous T1, T2*, and proton density quantification. Magn Reson
Med 2007;57:528–37 CrossRef Medline
3. Krauss W, GunnarssonM, Andersson T, et al. Accuracy and repro-
ducibility of a quantitative magnetic resonance imaging method
for concurrentmeasurements of tissue relaxation times andproton
density.Magn Reson Imaging 2015;33:584–91 CrossRef Medline
4. Warntjes J, Leinhard O, West J, et al. Rapid magnetic resonance
quantification on the brain: optimization for clinical usage.Magn
Reson Med 2008;60:320–29 CrossRef Medline
5. Blystad I, Warntjes J, Smedby O, et al. Synthetic MRI of the brain in
a clinical setting. Acta Radiol 2012;53:1158–63 CrossRef Medline
6. Granberg T, Uppman M, Hashim, et al. Clinical feasibility of syn-
thetic MRI in multiple sclerosis: a diagnostic and volumetric vali-
dation study. AJNR Am J Neuroradiol 2016;37:1023–29 CrossRef
Medline
7. Hagiwara A, Hori M, Suzuki M, et al. Contrast-enhanced synthetic
MRI for the detection of brainmetastases.Acta Radiol Open 2016;5:
2058460115626757 CrossRef Medline
8. Betts M, Leach J, Jones B, et al. Brain imaging with synthetic MR
in children: clinical quality assessment. Neuroradiology 2016;58:
1017–26 CrossRef Medline
9. Struck A, Westover M. Variability in clinical assessment of neuro-
imaging in temporal lobe epilepsy. Seizure 2015;30:132–35 CrossRef
Medline
10. Wattjes M,Wijburg M, Vennegoor A, et al.Diagnostic performance
of brainMRI in pharmacovigilance of natalizumab-treatedMS pa-
tients.Mult Scler 2016;22:1174–83 CrossRef Medline
11. Gulani V, Schmitt P, Griswold M. Towards a single-sequence neu-
rologic magnetic resonance imaging examination: multiple-con-
trast images from an IR TrueFISP experiment. Invest Radiol 2004;
39:767–74 CrossRef Medline
12. Osborn AG, Salzman KL, Barkovich AJ. Diagnostic Imaging: Brain.
2nd ed. Friesens: Amirsys; 2010
13. Krupa K, Bekiesin´ska-Figatowska M. Artifacts in magnetic reso-
nance imaging. Pol J Radiol 2015;80:93–106 CrossRef Medline
14. Ahn S, Park SH, Lee KH. How to demonstrate similarity by using
noninferiority and equivalence statistical testing in radiology re-
search. Radiology 2013;267:328–38 CrossRef Medline
15. Tha K, Terae S, Kudo K, et al.Differential diagnosis of hyperintense
cerebrospinal fluid on fluid-attenuated inversion recovery images
of the brain, Part II: non-pathological conditions. Br J Radiol 2009;
82:610–14 CrossRef Medline
16. Tha K, Terae S, Kudo K, et al.Differential diagnosis of hyperintense
cerebrospinal fluid on fluid-attenuated inversion recovery images
of the brain, Part I: pathological conditions. Br J Radiol 2009;82:
426–34 Medline
FIG6. Subdural hematomaonT2FLAIR in synthetic andconventional 3T
MR imagingdemonstrating pronounced artifacts. Conventional (left) and
synthetic (right) T2 FLAIR images are shown for a patient with subdural
hematoma, in which synthetic T2 FLAIR has notable granulated hyperin-
tensities and lacks contrast between the lesion and surrounding tissues.
Artifacts of this severity level were rare among synthetically recon-
structed images, possibly due to issues in theMDME acquisition that are
typically resolved on rescanning. For cases demonstrating these granu-
lated hyperintensities on the synthetic T2 FLAIR, artifacts were readily
recognizable by characteristic distortion and correlationwith other con-
trast views without apparent artifacts. While these could necessitate
rescanningwith conventional T2 FLAIR in somecases,whencoupledwith
other contrast views, these artifacts did not interferewith thediagnostic
accuracy of synthetic MR imaging.
AJNR Am J Neuroradiol ●:● ● 2017 www.ajnr.org 7
17. Ogbole G, SoneyeM, Okorie C, et al. Intraventricular cerebrospinal
fluid pulsation artifacts on low-field magnetic resonance imaging:
potential pitfall in diagnosis? Niger Med J 2016;57:59–63 CrossRef
Medline
18. HakkyM, Pandey S, Kwak E, et al.Application of basic physics prin-
ciples to clinical neuroradiology: differentiating artifacts from true
pathology on MRI. Am J Roentgenol 2013;201:369–77 CrossRef
Medline
19. Andre J, Bresnahan B,Mossa-BashaM, et al.Toward quantifying the
prevalence, severity, and cost associated with patient motion dur-
ing clinical MR examinations. J Am Coll Radiol 2015;12:689–95
CrossRef Medline
20. Dagher AP, Smirniotopoulos J. Tumefactive demyelinating lesions.
Neuroradiology 1996;38:560–65 CrossRef Medline
21. Mabray M, Cohen B, Villanueva-Meyer J, et al. Performance of appar-
ent diffusion coefficient values and conventionalMRI features in dif-
ferentiating tumefactive demyelinating lesions from primary brain
neoplasms.AJR Am J Roentgenol 2015;205:1075–85CrossRefMedline
22. Atighechi S, Zolfaghari A, Baradaranfar M, et al. Estimation of sen-
sitivity and specificity of brain magnetic resonance imaging and
single photon emission computed tomography in the diagnosis of
olfactory dysfunction after head traumas. Am J Rhinol Allergy 2013;
27:403–06 CrossRef Medline
23. Wollman D. Sensitivity and specificity of neuroimaging for the di-
agnosis of Alzheimer’s disease. Dialogues Clin Neurosci 2003;5:
89–99 Medline
24. Sabol Z, Resic´ B, Juraski R, et al.Clinical sensitivity and specificity of
multipleT2-hyperintensities onbrainmagnetic resonance imaging
in diagnosis of neurofibromatosis type 1 in children: diagnostic ac-
curacy study. Croat Med J 2011;52:488–96 CrossRef Medline
25. Polman C, Reingold S, Banwell B, et al. Diagnostic criteria for mul-
tiple sclerosis: 2010 revisions to theMcDonald criteria. Ann Neurol
2011;69:292–302 CrossRef Medline
26. ZaitsevM,Maclaren J,HerbstM.Motion artifacts inMRI: a complex
problem with many partial solutions. J Magn Reson Imaging 2015;
42:887–901 CrossRef Medline
8 Tanenbaum ● 2017 www.ajnr.org
	Synthetic MRI for Clinical Neuroimaging: Results of the Magnetic Resonance Image Compilation (MAGiC) Prospective, Multicenter, Multireader Trial
	MATERIALS AND METHODS
	Participants and Clinical Assessments
	Image Acquisition
	Radiologic Assessments
	Statistical Analysis
	RESULTS
	Overall Diagnostic Quality
	Legibility of Anatomic/Morphologic Features
	Artifacts Occurrence and Characterization
	Diagnostic Performance
	DISCUSSION
	CONCLUSIONS
	ACKNOWLEDGMENTS
	REFERENCES